CivArchive
    Preview 45091843
    Preview 45091878
    Preview 45091879
    Preview 45091880
    Preview 45092087
    Preview 45092436
    Preview 45092873
    Preview 45093435
    Preview 45093459
    Preview 45093818
    Preview 45094355
    Preview 45094737
    Preview 45095937

    Common advise

    Take the prompt from Training Data. Add it into your base prompt. Remove some tags which are redundant or incompatible with your idea. Use ADetailer or HiResFix x1.2. Give it 30 steps or higher to render.

    You can do 1st pass with Euler A sampler and high CFG (like 7). And then add micro contrast and grain using Heun sampler w/ Karras scheduler in 2nd pass like ADetailer.

    Features

    These LoRa's are made with my own landscape and sky photography. There are urban, industrial, nature and suburban themes in this series of LoRa.

    I would be glad for mentions and suggestions. This will help me to create new photographs and new LoRa's based on it.

    PS: BUZZ also good to add motivation ;)

    Description

    Features: down town passages and pathways, autumn colors, bad muddy roads, decaying fences, lead sky, yellow trees, dried grass, fog, mist, old weathered houses, dirty homeless white dog. Vivid yellows, earthy tones, dimmed out other colors.

    FAQ

    Comments (25)

    damarienezel296Dec 12, 2024
    CivitAI

    Do you live in Pripjat? or Bajkonur? I will test it.

    homoludens
    Author
    Dec 12, 2024· 1 reaction

    nope, but nature and mood is similar

    damarienezel296Dec 12, 2024

    @homoludens I will test it. Does it add realism to whole picture? I mean if i use a cartoon lora, will it be ultrarealistic?

    homoludens
    Author
    Dec 12, 2024

    @damarienezel296 It was trained on XenEro v1.7 which is almost fully realistic. I can expect that this LoRa won't convert Anime Pony into realistic.

    But I uploaded 70ths - 90ths Epochs which can be slightly over-trained and maybe it will rewrite visuals.

    I suggest you to install Lora Block Weights ext for A1111 and use a weights like
    <lora:...:0.2:0.8>
    Second sub-weight is for visual part of LoRa (OUT). First sub-weight is for IN blocks. If you need to test this LoRa for visuals (colors, details) you should start with lowest IN weight and high OUT weight.

    homoludens
    Author
    Dec 12, 2024

    Weights 0.8 is quite STRONG. As you can see with RiverSide LoRa, 0.8 will rewrite whole picture if you use a PONY style prompt from Training data as a part of your full prompt.

    damarienezel296Dec 12, 2024

    @homoludens I know, but i have a very good realistic setting with CyberPony, It works(ok realism) with cartoon loras. So if this doesnt change the faces too much, it can add even more realism. Or i will just simple use with highres fix.

    homoludens
    Author
    Dec 12, 2024· 1 reaction

    @damarienezel296 HiRes Fix or ADetailer should not influence the colors. They should affect details only. I hope this LoRa will add a realistic color palette in your setup. With XenEro and high OUT weight it gives a colors very close to what I saw through the Electronic Viewfinder of my Nikon Z7 (grass, yelow leaves, sky, building facades). This Lora laks micro details of textures IRL because CivitAI don't allow me to train in higher resolution. But in sense of composition and colors it is very realistic.
    Didn't test with cartoonish or ART CKPTs. Will be glad if you post some images with critics (tech, aesthetic issue details if any).

    damarienezel296Dec 12, 2024

    @homoludens I will

    homoludens
    Author
    Dec 12, 2024

    @damarienezel296 Rough test with Prefect Pony
    https://civitai.com/posts/10222097

    homoludens
    Author
    Dec 12, 2024

    @damarienezel296 Test with realistic PONY gives shitty result. I should train alternative strand of these models for CyberRealistic or for some other Pony models

    damarienezel296Dec 12, 2024· 1 reaction

    @homoludens Not at all shit!!!!!! Man! i uploaded 4 pics, flagged now. At 0.7 it gives the exact result, what i wante, no face change, only greyish amateur look, with more realistic skin.

    Haha, very good job.

    damarienezel296Dec 12, 2024

    @homoludens And with a full 3d lora.

    homoludens
    Author
    Dec 12, 2024

    @damarienezel296 results from CivitAI's generator was terrible to me. Maybe because of shitty sampler. Glad you have satisfying results.

    damarienezel296Dec 12, 2024

    @homoludens these mothefuckers flags everything. I uploaded SFW girl, and still flagged, she is too sexy.

    homoludens
    Author
    Dec 12, 2024

    @damarienezel296 initial flag from autoflagger reacts on several words and overall realism. Avoid using girl, 1girl, little, petite, virgin, minor, shit, feces, etc.

    damarienezel296Dec 12, 2024· 1 reaction

    @homoludens none of it, not even real life person, and sfw, and has breast

    Ok, i made her face young, to get likeness, but thats it.

    damarienezel296Dec 12, 2024

    @homoludens all in all, it doesnt matter, S-tier lora, thanx.

    homoludens
    Author
    Dec 12, 2024

    @damarienezel296 Maybe you should add an eye bags to make it older?

    damarienezel296Dec 12, 2024

    @homoludens Haha, no i made her a bimbo granny, a real 25-35 looking whore, still flagged.

    homoludens
    Author
    Dec 12, 2024

    @damarienezel296 make a pure experiment. Prompt a "puritanical concrete cube, mormon adept"

    homoludens
    Author
    Dec 12, 2024

    @damarienezel296 check for bad simpols appearing some times when you copy past some prompts:

    " |“"

    damarienezel296Dec 12, 2024

    @homoludens puritanical concrete cube, mormon adept. Does it do something?

    homoludens
    Author
    Dec 12, 2024

    @damarienezel296 it was just a intuitive suggestion. Use only these words which should no trigger the NSFW flagging. If you will still get a flag then it must be something on account level. Maybe there is some flag on whole account which forces the review no mater what content you post.

    damarienezel296Dec 12, 2024

    @homoludens I wanted to show you, it works even with 3d loras, no uploaded Harley Quinn, flagged, they are trolling me. they know my setup, i always use that background. On the week, i have a chat with mods, to not harrash a user. I harrashed that retard, because, he keeps his setting, no prompt and other settings on his loras. So i harrashed a bit that retard. Im now on VPN, so there is only one thing left, they know my setup:))

    damarienezel296Dec 12, 2024

    @homoludens Hahaha, i was right, without background, its there, idiots:)))

    LORA
    Pony

    Details

    Downloads
    384
    Platform
    CivitAI
    Platform Status
    Available
    Created
    12/12/2024
    Updated
    6/12/2026
    Deleted
    -
    Trigger Words:
    look into training data

    Files

    Autumn_Vibes_set1_196.safetensors

    Mirrors

    captions.zip

    Mirrors

    CivitAI (1 mirrors)

    Available On (1 platform)

    Same model published on other platforms. May have additional downloads or version variants.